Closed Jacob1010-h closed 8 months ago
The issue we ran into last night seems to be due to the sheer size of the dataset we are dealing with here. I've found a different way to do it, and would like to show it to you @YoungChadawan later today
I'm on the right track, but more will be done later today, there are some bugs with the python code right now which I need to look over. However once I get it working, this should be a very good ai model 😃 (9,000+ images training, and abt 25,000+ total)
Currently an issue with tenserflow updating their stuff to version 2 https://colab.research.google.com/drive/1adWIOWY9Ga_uUUq92FBnsvlC2uAqBjp7?usp=sharing
Running into errors using your linked colab on exporting the trained inference graph. Started using https://colab.research.google.com/github/EdjeElectronics/TensorFlow-Lite-Object-Detection-on-Android-and-Raspberry-Pi/blob/master/Train_TFLite2_Object_Detction_Model.ipynb#scrollTo=tQTfZChVzzpZ which has an outdated TensorFlow. Going back to your linked one however it is not detecting the label map for some reason will have to figure out why
Having trouble training the model. Keep getting the error "ModuleNotFoundError: No module named 'tensorflow_io'". I believe this has something to do with the dataset URLs I am giving it in prepare data. As of now I have given both the COCO link but it does say to do Tensorflow record for the first and COCO for the 2nd. However, that does not seem to work either.
We are going to go back to a working dataset for the notes right now, and then when something is updated with the tflite models then we will take a look
Then we export to TF lite!!
Took them long enough
me when i brick my limelight
This includes Tuning and proper calibration with updated data